On the cutoff frequency of clarinet-like instruments.

Geometrical versus acoustical regularity

Elise Moers, Jean Kergomard

To cite this version:

Elise Moers, Jean Kergomard. On the cutoff frequency of clarinet-like instruments. Geometrical
versus acoustical regularity. Acta Acustica united with Acustica, Hirzel Verlag, 2011, 97, pp.984-996.
�10.3813/AAA.918480�. �hal-00559014v2�

HAL Id: hal-00559014

https://hal.archives-ouvertes.fr/hal-00559014v2
Submitted on 18 Aug 2011

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci-
entific research documents, whether they are pub-
lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
 On the cutoff frequency of clarinet-like
instruments
Geometrical vs acoustical regularity
E. Moers, J.Kergomard
Laboratoire de Mécanique et d’Acoustique, CNRS UPR 7051, 31 Chemin
Joseph Aiguier, 13402 Marseille Cedex 20, France

Abstract
A characteristic of woodwind instruments is the cutoff frequency of
their tone-hole lattice. Benade proposed a practical definition using the
measurement of the input impedance, for which at least two frequency
bands appear. The first one is a stop band, while the second one is a pass
band. The value of this frequency, which is a global quantity, depends
on the whole geometry of the instrument, but is rather independent of
the fingering. This seems to justify the consideration of a woodwind with
several open holes as a periodic lattice. However the holes on a clarinet
are very irregular. The paper investigates the question of the acoustical
regularity: an acoustically regular lattice of tone holes is defined as a
lattice built with T-shaped cells of equal eigenfrequencies. Then the paper
discusses the possibility of division of a real lattice into cells of equal
eigenfrequencies. It is shown that it is not straightforward but possible,
explaining the apparent paradox of Benade’s theory. When considering
the open holes from the input of the instrument to its output, the spacings
between holes are enlarged together with their radii: this explains the
relative constancy of the eigenfrequencies.
PACS: 43.75

1 Introduction

In his paper of 1960, Benade [1] proposed to use the theory of periodic media
in order to analyze the effects of a row of tone-holes of wind instruments. He
was mainly interested in the length correction at the input of a regular lattice of
holes, when they are either all closed or all open. He discovered the existence of
an important frequency, the cutoff frequency of the lattice of open holes. Below
cutoff, at low frequency (in the stop band), a wave is evanescent, i.e. exponen-
tially decreasing, while above cutoff (in the pass band), it can propagate.

Later, in his book [2], he gave many details about this frequency for real
instruments and published experimental results showing the relative indepen-
dence of this frequency with respect to the fingerings (except the fork ones) in
the first register of oboes, bassoons and clarinets. In addition he explained how

1
this frequency is correlated to the tone-color adjectives used by musicians to
describe the overall tone of an instrument (see [1] p. 486).
For a perfectly periodic lattice, the cutoff frequency is independent of the fin-
gering. For real instruments the small variation of the cutoff with respect to
the fingering suggests a great regularity of the tone-hole lattice. However this
seems to be in contradiction with the great irregularity of the holes of a clar-
inet. This apparent paradox is the basis of the motivation of the present paper,
and was already noticed by Benade in a posthumous article [3]. Comparing 2
clarinets, one with a regular tone hole lattice and an ordinary one, he stated
that the input impedances, measured at low level, “were almost identical. This
was of course interesting and happy news, because it helped justify the use of
formal mathematical physics for a slowly varying lattice on a geometrically quite
irregular physical structure”.
Benade proposed a practical method of determination of this frequency based
upon the measurement of the input impedance. Two examples will be shown
in Figs. 2 and 3. In general, this works properly even for a small number
of open holes, thanks to a rather clear distinction between the stop band and
the pass band. In principle the measured quantity is a global quantity, related
to the input impedance, which a priori depends on the whole geometry of the
instrument for a given fingering, i.e. for a given configuration of open holes.
It a priori depends therefore on the hole irregularity and the termination of
the instrument. As a consequence, the relative independence with respect to
fingering is not obvious. Notice that using the measurement of any transfer
function, the effect of a (global) cutoff frequency strictly appears only for a
periodic, infinite and lossless lattice. When losses exist, the definition is less
strict but precise. When the lattice is of finite length or/and irregular, the
definition of a global cutoff remains possible in general, as it will be discussed in
the present paper. In what follows, we will define the global cutoff frequency as
the frequency separating two frequency bands, as viewed on the input impedance
curve. It is possible to find an analogy with horn theory: above the global
cutoff, the input impedance curve has no resonances.
In his book, Benade also discussed the effect of irregularity. Let us cite him
(p. 449): “If the lattice is irregular, theory shows that: (1) if the first and
second open-hole segments of the lattice (taken by themselves) have widely dif-
ferent cutoff frequencies, the observed value of fc for the composite system has
an intermediate value for its cutoff frequency; and (2) at the lower frequencies,
the properties of the first segment still dominate the implications of fc .” We
can remark that here Benade regards the cutoff frequency as a local quantity,
defined for one segment and not for a complete lattice. We will define the local
cutoff frequency as the frequency calculated from a given cell (or segment), cor-
responding to the theoretical cutoff of the periodic medium built with an infinity
of cells identical to the considered one. In a paper on cutoff frequencies of flutes,
Wolfe and Smith [4] also implicitly considered a local definition (“the cutoff fre-
quency varies from hole to hole”, see Figure 4 of the paper) and calculated it
using “typical values” for the dimensions and spacing of holes, in order to use
the formula corresponding to a periodic medium. In addition they calculated
the deviation of the calculated frequencies, exhibiting a large value for it. (For
the classical flute, only the tone holes used in the diatonic scale were included
in the means. For the modern instrument, all holes except the trill holes were
included).

2
 For a perfectly periodic lattice, i.e. a perfectly geometrically regular lattice,
there is no difficulty in defining a local cutoff, because the lattice is divided into
identical cells, one cell determining a cutoff. For that case, local and global
cutoff frequencies coincide, at least when the lattice is long enough, and the
measured global cutoff does not depend on the fingering.
Two questions are examined in the present paper concerning a real instru-
ment, with irregular geometry:

1. Is it possible to define an “acoustical regularity”1, for which coincidence

between local and global cutoff frequencies is possible even without strict
periodicity? We will prove that the answer is positive for a lattice of
holes, at least at low frequencies, under the condition that the local cutoff
frequency is uniform and the lattice long enough. It will be explained how
is possible to build an instrument with this property. This question can
be regarded as a direct problem.
2. Concerning the inverse problem, starting from the result of Benade that
the global cutoff depends little on the fingering, and on the number of open
holes, is it possible to find an “acoustical regularity” in a real clarinet?
The answer will be shown to be positive. For this purpose, the delicate
problem of the division of a real clarinet into acoustically regular cells
needs to be solved. The solution is not simple, as will be shown hereafter.

The final objective of the paper is to compare the obtained local cutoffs
to the global ones, where the local cutoffs are calculated from the geometrical
dimensions of different holes, while global cutoffs are measured for different
fingerings with one or several open holes (for this purpose, it will be be necessary
to extend the previous definition of the local cutoff). In general the lattices
are with losses and sometimes of short length, and in addition they are not
regular, thus the definition of the global cutoff using the input impedance is
necessarily done with a non negligible uncertainty. In other words, there is
no perfect separation between two frequency bands. Therefore for the sought
comparison, an accurate calculation of the local cutoff is not useful and would
be illusory. This allows an approximate treatment, with simplification of the
model. Nevertheless this leads to a satisfactory comparison with experiment.
The outline of the paper is as follows: section 2 reviews the theory of Benade,
and adds a useful interpretation of the cutoff frequency as the eigenfrequency
of a cell (i.e. a segment) of the lattice. Section 3 presents experimental results
for a Yamaha Y250 clarinet for the global cutoff frequencies measured from the
input impedance, confirming Benade’s results. In addition numerical simulation
exhibits the important effect of the termination, even for a periodic lattice.
Section 4 discusses the first above-mentioned question. Section 5 tries to answer
the second question: in a first step, the tube is supposed to be cylindrical and
without closed tone holes, and in a second step some corrections are sought to
this simplification. In section 6 global and local cutoff frequencies of the clarinet
are compared and acoustical regularity is discussed.
1 Notice that in Ref. [3], Benade wrote “Acoustical regularity is a virtue”, but the signifi-

cance of this expression was different: it was related to the homogeneity of input impedance
for the different fingerings.

3
2 Periodic lossless lattice of holes: the cutoff
frequency and its interpretation as an eigen-
frequency
Benade [2] proposed a formula for the first cutoff frequency of a perfectly pe-
riodic lattice of open holes, valid at low frequencies (it is recalled below as Eq.
(6)). We remark that the corresponding frequency is the eigenfrequency of a
Helmholtz resonator built as follows: the volume is that of a portion of the
main tube with a length equal to the spacing between two adjacent holes, and
the neck is the open tone hole. In this section we review the basic model, and
explain why this remark is true, even at higher frequencies. The considered
lattice is built with a row of T -shaped cells (see Fig.1).
Let us consider the classical transfer-matrix description of a symmetrical
cell, relating pressures pn and flow rates un at the extremities (with indices n
and n + 1), as follows:
    
pn A(ω) B(ω) pn+1
= . (1)
un C(ω) D(ω) un+1
The transfer matrix has the following properties: coefficients A and D are equal,
because of symmetry, and the determinant is unity, because of reciprocity. We
ignore losses, thus A is real, and B and C imaginary (in what follows both visco-
thermal and radiation losses are ignored). In an infinite lattice, the travelling
waves are
pn = p0 exp(±nΓ)
un = u0 exp(±n) ,
where Γ is the propagation constant. Using Eq. (2) for the two cells (n − 1,
n) and (n, n + 1) leads to cosh Γ = A. Therefore at cutoff, A = ±1, Γ = 0 or
π. At the cutoff frequency, the complex amplitude of a traveling lattice wave of
pressure (respectively of flow rate), is constant from one cell to the next one,
with a factor ±1. For symmetry reasons detailed hereafter, it can be deduced
that either the flow rate or the pressure vanishes at the extremities of the cells.
The relationship between pressure and flow rate waves is defined by the
characteristic admittance Yc (or impedance Zc = 1/Yc ), given by:
sinh Γ A2 − 1 C C2
Yc = thus Yc2 = 2
= = 2 . (2)
B B B A −1
When Γ is imaginary (and Zc real), waves propagate, while if Γ is real (and Zc
imaginary), waves are evanescent. The first case corresponds to pass bands, the
second one to stop bands.
At cutoff, A = ±1, thus BC = 0, i.e. either B or C vanishes. If C vanishes,
Yc vanishes too. The flow rate being proportional to Yc for the waves in the two
directions, it is zero for any value of n, for both an infinite or a finite lattice.
Therefore the pressure wave is constant:
pn = Apn+1 = ±pn+1 .
If pn = pn+1 , the pressure field is symmetrical, while if pn = −pn+1 , the pressure
field is antisymmetrical. The dual situation, reversing the roles of pressure and

4
Figure 1: Two T -shaped cells of a lattice of tone holes. A T-shaped cell consists
of the tone hole and a length ℓ of the main tube on both sides of the hole. The
cross section areas are S = πa2 and Sh = πb2 , where a and b are the radii of
the main tube and the hole, respectively.

flow rate, occurs if B = 0. Finally the cutoff frequencies are the eigenfrequencies
of a cell with either Neumann (un = 0, if C = 0) or Dirichlet (pn = 0, if B = 0)
conditions at the extremities.
The next question is the distinction between the first eigenfrequencies sat-
isfying the termination conditions. At low frequencies, for a cell of a tone-hole
lattice, the coefficient A is larger than unity , therefore the waves are evanescent
and the first cutoff occurs for A = +1, i.e. when either the pressure field or the
flow rate field is symmetrical. As it is well known, a Helmholtz resonator has
an eigenfrequency very low when its volume is closed, therefore the first cutoff
frequency of the lattice, which is the subject of the present paper, corresponds
to the Neumann boundary conditions, with a symmetrical pressure field. This

5
is true even if the wavelength at cutoff is not larger than the dimensions of a
cell. In Ref. [5] expressions are given for the four types of cutoff frequencies (see
also Appendix B of the present paper), with a comparison of the first four ones,
confirming the fact that the lowest cutoff is that of the Helmholtz resonance of
a T-shaped cell. It corresponds to the condition Yc = 0, or C = 0, and this is
in accordance with a mathematical analysis done by Benade (in his Eq. 8, the
cutoff is obtained when the denominator vanishes [1]). To our knowledge, this
interpretation is new.
It can be concluded that considering the transfer matrix or the equivalent
circuit of a tone-hole (see Refs [6, 7]), the element corresponding to the anti-
symmetrical field, which is a series impedance denoted Za , does not appear in
the expression of the first cutoff frequency (this is a rigorous result, without any
approximation). Nevertheless ignoring this series impedance is not valid at any
frequency. In Appendix B, it is shown how this series impedance could be taken
into account in order to generalize the present approach, but for the purpose of
the paper, simplified models are sufficient and we ignore it.
Now, if the height of the hole chimney is assumed to be much shorter than
the wavelength, and if losses are ignored, the effect of the tone hole is reduced
to a shunt acoustic mass, denoted mh . The coefficients for transfer matrices of
a T −shaped cell are given by standard acoustic theory:

A = D = 1 − 2 sin2 kℓ + jY zc sin kℓ cos kℓ (3)

zc 2j sin kℓ cos kℓ − Y zc sin2 kℓ
 
B =
zc−1 2j sin kℓ cos kℓ + Y zc cos2 kℓ ,
 
C =

where Y = (jωmh )−1 is the shunt admittance of the hole and zc = ρc/S, the
characteristic impedance of the main tube. ρ is the air density, c the sound
speed, S = πa2 the cross-section area of the tube, assumed to be cylindrical,
2ℓ the spacing between two holes, ω the angular frequency, and k = ω/c the
wavenumber.
The equation satisfied by the cutoff frequency is given by C = 0:
ρc
j cot kℓ = 2jωmh . (4)
S
The left-side member is the impedance on both sides of the tone hole, deduced
from the Neumann boundary condition, while the right-side member is twice
the shunt impedance of the tone hole. The mass mh is the sum of the mass of
the planar mode in the hole and of the masses corresponding to the radiation
impedance into surrounding space and into the main tube. It is approximately
equal to:
mh = ρht /Sh , with ht ≃ h + 1.6b, (5)
where h is the height of the hole, and Sh = πb2 the cross section area of the
hole. Notice that ht is denoted te by Benade.
Solving Eq. (4) when kℓ << 1 leads to the result (Ref. [2]):
c 1 1 c b 1
fc = p = √ . (6)
2π ℓ 2mh /m 2π a 2ℓht

where m = ρℓ/S is the acoustic mass of the portion of the main tube of length
ℓ (notice that in Eq. (6) the compressibility of air in the main tube appears, via

6
the acoustic compliance Ca = ℓS/ρc2 , but its inertia does not). The exact value
of ht depends on several parameters, such as the undercutting of the hole or the
existence of a key pad, but this is not critical for the present study, especially
because the cutoff depends on the square root of this mass.
A better approximation for the solution of Eq. (4) is the following (see
Refs [8, 9]):
c 1 1 c
fc = = . (7)
2π ℓ 2mh /m + 1/3
p p
2πℓ 2(a/b)2 ht /ℓ + 1/3

It is obtained by expanding cot kℓ to the next order in k 2 ℓ2 . This gives a

condition of validity of Eq. (6):

b2 ht m
<< 6 or << 6 , (8)
a2 ℓ mh

If kc = 2πfc /c, this condition is equivalent to: kc2 ℓ2 << 1 (the half spacing
between two holes is much smaller than the cutoff wavelength, i.e. the elements
of the system are lumped). For a clarinet, typical values of the cutoff frequency
and length ℓ are 1500 Hz and 10 mm, thus kc ℓ ∼ 0.3, and the condition is
satisfied.

The dimensions and locations of the holes are given in Appendix A (Table
1). Table 2 of the same appendix indicates the first opened holes for the different
fingerings.

3 Determination of global cutoff frequencies

Benade [2] proposed a simple method to measure the “global” cutoff frequency
of a given instrument, considering the curve of the input impedance modulus.
The first two frequency bands generally appear rather clearly: the first one
with high and regular peaks, the second one with smaller and irregular peaks.
Benade defined the cutoff as the boundary between the two bands. As stated
in the introduction, in principle this method is perfect for a perfectly periodic
lattice (i.e. regular, lossless and infinite). What happens for a real lattice is
discussed hereafter.
The explanation given by Benade (p. 434) is based upon the strong radiation
of the holes above cutoff. This can be more detailed: below cutoff, in the stop
band (waves are evanescent), the effective length of the tube is very close to the
tube cut at the first open hole, thus the frequency interval between resonances
is large. Moreover boundary layer losses (which are preponderant in this range)
are small, thus the impedance peaks are high.
On the contrary, above cutoff (in the pass band), the effective tube is divided
into two portions. The first one is without open holes, while the second one
is the open-hole lattice. The phase velocity and characteristic impedance are
different in the two portions, thus the impedance peaks are irregular. Moreover
boundary-layer losses exist over a large length, and several open holes radiate
efficiently. Therefore the peaks are lower than in the stop band. The efficient
radiation is due not only to the number of holes radiating, but also to the
external interaction between the holes, as shown in Ref. [10].

7
 Benade [2] measured modern and baroque instruments, and found that
baroque instruments have lower cutoff frequencies than modern ones. The ex-
planation seems to be evident thanks to the analysis of the previous section:
first of all, the holes of baroque instruments are generally narrower than those
of modern instruments. In addition baroque instruments are basically diatonic
instruments while roughly speaking the basis of modern instruments is more
chromatic. Therefore spacings between open holes are larger for baroque in-
struments than for modern ones. These two facts with the interpretation of the
cutoff frequency as the eigenfrequency of a cell viewed as a resonator explain the
differences in cutoff frequencies. A consequence is the slightly wider compass of
modern instruments, even if it is possible to play notes with frequencies higher
than the cutoff (obviously a complementary explanation for the wider compass
of modern instruments is the addition of new holes). Otherwise the question of
the influence of the cutoff frequency on the sound spectrum has been discussed
rather rarely, but Benade and Kouzoupis can be cited [11], as well as Ref. [4] for
the flutes. This question is out of the scope of the present paper. Finally the
question of the influence of cutoff on directivity of woodwinds has been treated
in Refs. [1, 8, 10, 12].

3.1 Measurement results

30
D4
20 fc = 1450 Hz
10
|Z| [dB]

−10

−20

−30

−40
500 1000 1500 2000 2500 3000

180
D4
Arg(Z) [Degrees]

90 f = 1450 Hz
c

−90

−180
500 1000 1500 2000 2500 3000
Frequency [Hz]

Figure 2: Measured impedance curve for note D4. The cutoff fc is found
to be 1450. The scale for the impedance modulus is the logarithmic one:
20 log(|Z|/zc ).

Benade deduced the cutoff frequency from the measured modulus of the
input impedance on a linear scale. Actually it is easier in practice to use either
the modulus of the impedance on a logarithmic scale or its argument. This is

8
 30
f = 1130 Hz A3
c
20

10
|Z| [dB] 0

−10

−20

−30

−40
0 500 1000 1500 2000 2500 3000

180
A3
f = 1130 Hz
Arg(Z) [Degrees]

90 c

−90

−180
0 500 1000 1500 2000 2500 3000
Frequency [Hz]

Figure 3: Measured impedance curve for note A3. The cutoff fc is found to
be 1130 Hz. The scale for the impedance modulus is the logarithmic one:
20 log(|Z|/zc ).

often better because a slope inversion clearly appears for almost every fingering,
even when only a small number of holes is open. This leads to a definition of
the global cutoff with a precision in practice better than 1%. Nevertheless, as
stated in the introduction, this does not mean that the separation of the two
frequency bands is precise, the definition being somewhat arbitrary. A further
discussion is given in [13].
Figures 2 and 3 show two examples of measured input impedance curves
for the notes D4 and A3. For the note D4 (262 Hz), the global cutoff is found
to be 1450 Hz, while for the note A3 it is 1130 Hz. For all the notes of the first
register, the measurement of the cutoff frequency is easy, even when only one
hole is open (note F3): Benade and Kouzoupis [11] explained this fact by the
effect of the bell, “which serves as a more or less surrogate for an open-hole
lattice” (sect VIID, see also Ref. [3]).
An interesting result is that the cutoff does not vary very much (a variation
of 20% is much smaller than the variations of geometrical dimensions), even
for this kind of notes, as it will be seen on Figure 4, which shows the results
for the different notes of the clarinet studied (Yamaha Y250). The results are
within the range of results obtained by Benade, who measured several different
instruments. Notice that Benade gave results for the first register, for the same
notes as those we have studied, except the notes for which the first open tone
hole is provided with what we call a “closed key”, i.e. a key open for this note
only (see Appendix A Table 2). Otherwise, as expected, the cutoff frequencies
for the second register, when the register hole is open, are very close to the
cutoffs for the first register, for the corresponding fingering.
For the first register, two groups of notes can be observed, above and below

9
B4, around 1450 Hz and 1150 Hz, respectively. We will see in section 3.2 that
this difference is not related to regularity, but it is due to the termination effect.

2000

1800

1600
Cutoff frequency [Hz]

1400

1200

1000

800

600

400

200

0
E6 C#6 A#5 G5 E5 C#5 A#4 G4 E4 C#4 A#3 G3 E3

Figure 4: Cutoff frequencies measured for the Yamaha Y250 clarinet. The graph
covers all fingerings, and for the first register exhibits a great similarity with
the results of Benade, the scale being chosen to be similar.

The device used for the impedance measurement is based upon the mea-
surement of acoustic pressures in two cavities separated by a flow rate source,
marketed by CTTM [14].

3.2 Simulation results

In order to get more insight into the problem, some simple simulations using
transfer matrices like Eq. (3) have been carried out. No series impedances are
taken into account (see Eqs. (3)), but boundary layer losses and radiation, given
by standard theory, are considered. Radiation takes external interaction into
account, via a global admittance matrix (see Ref. [10]). A simplified shape of the
bell is used. Some open holes are considered through a unique, equivalent tone
hole, as explained hereafter in section 5.2, resulting in a lattice with 11 open
holes. The (global) cutoff frequencies are deduced using the input impedance
curve.
Fig.5 shows an interesting qualitative agreement between this model and
experiment, sufficient for our purpose. The discrepancy is less than 11%, this
value occurring for the note A♯3. This fact can be related to the location of the
limit between the two groups of values, near to A♯3.
The most interesting result is the comparison between the values for the
geometrical data of the studied clarinet and those for a perfectly periodic lat-
tice, having 11 open holes with constant spacing 2ℓ = 0.0341m and theoretical
cutoff 1450 Hz, the radius being 7.5 mm. The total length is the same. The

10
 2000

1800
(2a)
1600

1400

1200
(1)
f
c
1000

(2b)
800

600

400

200

0
A4 G#4 G4 F4 E4 D4 C4 A#3 A3 G3

Figure 5: Measured and calculated global cutoff frequencies. Some fingerings

are not considered, see explanation in table 2 in Appendix A; moreover the two
lowest notes, E3 and F3, are not considered as well, only notes with at least two
open holes being considered. Solid, thick line: experimental results (also shown
in the previous figure). Dashdot line: numerical result of the simplified model
with a bell (see section 5.2.2). Solid, thin lines: numerical results for a purely
periodic lattice: (1) with a bell, (2) with a cylindrical tube replacing the bell
(for low notes, two possible values of the cutoff are shown, denoted 2a and 2b).
The thin, dotted line indicates the theoretical cutoff of the periodic lattice.

common value of the reduced masses, mh /ρ = 340 m−1 , is deduced from Eq.
(6). Notice that the exact value (Eq. (4)) of the theoretical cutoff is 1402 Hz;
the approximation (7), giving also 1402 Hz, is excellent.
The main features are the following:

• The differences between the results for the purely periodic lattice and the
simplified model of the irregular, real one are less than 5%; this can be
seen as a first indication of the existence of an acoustic regularity;
• The existence of two groups of values with a limit for A3 is roughly similar
for the two lattices;
• Above A3 (first group), when many holes are open, the global cutoff is
higher than the theoretical value (1402 Hz), some values being higher than
1500 Hz, the average being 6% higher than the theoretical one. Even for
the highest note, when all the holes are open, the global cutoff is 2% higher
than the theoretical one.
• The existence of two groups is due to the effect of termination only. This
can be checked by replacing the bell by a cylindrical tube of same length
and input radius. The values corresponding to the lowest notes (2nd group)
are strongly modified. The above cited sentence by Benade and Kouzoupis

11
 is probably true, because for the lowest notes, the determination of the
global cutoff is uncertain. For the cylindrical termination, irregular peaks
are found around 900 Hz, but the shape of the impedance curves differs
strongly from the typical curves shown in Figures 2 and 3.

Figure 6 shows a comparison of theoretical results for two kinds of lat-

tices: the one of the considered clarinet, and the above mentioned periodic
lattice2 bell). .

40

20

0
|Zin/Zc|[dB]

−20

−40

−60
0 500 1000 1500 2000 2500 3000

100

50
arg(Z )
in
0

−50

−100
0 500 1000 1500 2000 2500 3000
Frequency [Hz]

Figure 6: Calculated input impedance curve for the fingering D4. Solid line:
simplified model of the clarinet. Dashdot line: periodic lattice of the same
length and termination. .

For this particular case, it appears that the practical definition of the global
cutoff is easier for the real (irregular) lattice than for periodic one with the
same termination. It is a confirmation that the definition of the global cutoff is
not always easy in practice (For an infinite lattice without losses, the argument
should exhibit a discontinuity between the two bands).

4 Construction of an acoustically regular lattice

This section investigates if acoustically regular lattices can exist. It is known

since Anderson [15] that in a onedimensional medium, the effect of an infinite
number of random irregularities is the suppression of pass bands, and therefore
of cutoff frequencies. Obviously this asymptotic property cannot be observed
on musical instruments, because of their limited length. Moreover the theorem
2 between the first resonance frequencies comes from the difference in length of the tubes

upstream from the first open tone hole. The spacing between the first tone holes is significantly
smaller than the mean spacing, used for the periodic lattice. This will be seen in Fig. 8.

12
of Fürstenberg [16] concerning the product of random matrices indicates that
some exceptions to the Anderson’s result can exist. In Ref. [17], it is shown that
the product of matrices having the same characteristic impedance Zc is such an
exception. As a matter of fact, for that case:
   
N cosh Γi Zc sinh Γi cosh σ Zc sinh σ
Π = (9)
i Zc−1 sinh Γi cosh Γi Zc−1 sinh σ cosh σ
N
where σ = ΣΓ i .
i

The behavior is similar to this of the regular medium with the same total
propagation constant σ = nΓ. As a consequence, if a lattice is built with
irregular cells having the same characteristic impedance at every frequency, its
behavior is the same as the behavior of a perfectly periodic medium. If this
situation can exist for wind instruments, acoustic regularity can exist without
geometrical regularity. In particular stop and pass bands can exist: when σ is
imaginary (and Zc real), waves propagate, while if σ is real (and Zc imaginary),
waves are evanescent. This is now investigated.
The cutoff of a T -shaped cell is given by A = 1 or C = 0 (see Eq. (4)). In
order to exhibit the value of the cutoff wavenumber kc , using Eq. (4) for ω = ωc
and k = kc , the mass mh can be eliminated and the admittance Y = (jωmh )−1
is rewritten as follows:
kc
Y = −2jzc−1 tan kc ℓ .
k
Thus, using the definition of the transfer matrix (Eq. (3)):

B 1 + kkc tan kc ℓ tan kℓ

Zc2 = = zc2 . (10)
C 1 − kkc tan kc ℓ cot kℓ

The characteristic impedance is written with respect to two quantities, the half-
length ℓ of a cell and the cutoff wavenumber kc , the latter parameter replacing
the hole mass. If a lattice is built with cells of identical kc , the characteristic
impedance can be identical (i.e. independent of ℓ) at low frequencies if both kc ℓ
and kℓ are small quantities:

B 1 + O(kc2 ℓ2 )
Zc = = zc2 . (11)
C 1 − kc2 /k 2

The propagation constant is given by sinh2 Γ = BC with

BC = −4 sin2 kℓ cos2 kℓ (1 + kc k −1 tan kc ℓ tan kℓ)(1 + kc k −1 tan kc ℓ / tan kℓ)

(12)
thus with the same approximation,
r
k2
Γ = 2jϕℓ with ϕ = k 1 − c2 . (13)
k
ϕ is an equivalent wavenumber. Therefore, in the frequency range where the
cell dimensions are smaller than wavelength (and consequently where Eq. (6)
is valid), it is possible to build an acoustically regular lattice, provided that
the cutoff frequency of the cells is a constant. The length of the cells can be

13
arbitrarily chosen but cannot be too long, and for each cell the value of the
hole acoustic mass is deduced from the chosen value of the cutoff frequency.
Notice that starting from the input of the tube, the distance between holes can
either increase or decrease, and consequently the hole masses can decrease or
increase, respectively (e.g. the hole radii increase or decrease). The property of
the considered lattice is identical to that of a purely periodic one, therefore the
global cutoff is the same as the local cutoff of the cells. This is the answer to
the first question in the introduction.
Eqs. (11) and (13) suggest an analogy with the problem of an exponential
horn: this is discussed in Ref. [5]. Studying the “horn function” of a given bell
as Benade did, is equivalent to studying the local cutoff frequency of the bell,
which has strong variation for instance for a trumpet bell [18].
This result concerning acoustical regularity will be more complicated if the
antisymmetrical (negative) masses are taken into account in the calculation,
but the answer remains positive (see Appendix B). Other complications of the
model are possible if they can be compatible with the basic model, based upon
the association of an acoustic mass with an acoustic compliance. Fortunately
this is in general the case at low frequencies.

5 The inverse problem; analysis of an irregular

lattice
5.1 Statement of the problem
We will now analyze the lattice of a real instrument. Solving the inverse problem,
i.e. dividing a given lattice into cells having the same cutoff frequency, is not
an easy task, because the solution either is not unique or does not exist, as
explained hereafter. Obviously a first requirement for a method of division is
to re-obtain the initial division when considering a lattice built as explained in
section 4, instead of a real one.
We first assume the clarinet to be a purely cylindrical tube with holes of
different sizes. As it is known, the radius of the main tube does not vary very
much for a clarinet. We will see how it is possible to take into account conicity
of some portions as corrections. Another hypothesis has been made: in a first
step the closed holes have no influence, and their effect is taken as another kind
of correction.
For the present purpose, we consider 14 open holes. In three cases (holes
number 8 and 9, 14 and 15, 18 and 19, see Appendix A, table 2), two closely
spaced holes are opened simultaneously in order to get a given note: according
to the basic hypothesis of long wavelength, we choose to replace them by a single
hole, with a mass equivalent to the two masses in parallel, and located at the
middle of the interval of the two holes. It remains 11 equivalent holes, therefore
11 cells. All the other holes are closed.
Three methods of analysis have been investigated, the first two being based
upon different choices of division of the bore into cells, the third trying to extend
the definition of a local cutoff frequency without any division.

14
5.2 Methods of analysis and results
5.2.1 Division into symmetrical cells with varying eigenfrequency
A first method is implemented to divide the tone-hole lattice into T -shaped,
symmetrical cells. It is not possible to fix a common value for the cutoff fre-
quency, because doing that all the cell lengths become fixed (they are deduced
from the values of the hole masses and of the eigenfrequencies), and the cells
either will have overlap or do not re-build the complete lattice.
Thus for the chosen method no value for the cutoff is a priori fixed. An
initial parameter is arbitrarily chosen, i.e. for instance the half length of the first
(upper) cell in the lattice, denoted ℓ1 . Using iteration, this implies the length
of each cell, therefore the whole division of the lattice. The cutoff frequencies
of each cell can be deduced. They are a priori different, and depend on the
chosen value for ℓ1 . The ratio Rc of the cutoff highest value to the lowest one
is calculated, and the final choice of the parameter ℓ1 is found by searching for
the minimum value of this ratio, which appears very clearly.
The method is first tested on an ideal (acoustically regular) lattice of 11
tone-holes, as defined in the previous section. As expected, the minimum ratio
Rc is unity, for a length ℓ1 equal to the half length of the first cell. The method is
therefore capable of dividing correctly a lattice built to be acoustically regular.
On the contrary, when applied to the lattice of a clarinet, no division has
been found, because negative spacings between holes appear. This is probably
due to the close location of the two first upper holes. When ignoring these
two holes, a result is found, but the value of the minimum ratio is 2.8: this
value is high, while the other approaches, as it will be discussed hereafter, give
much smaller values, i.e. a much more uniform value for the cutoff frequencies.
Therefore this method has been abandoned [19].

5.2.2 Division into asymmetrical cells with constant eigenfrequency

For the second method, a more general model of acoustically regular lattices is
investigated, with one degree of freedom more. Asymmetrical cells are consid-
ered: the hole is not necessarily located at the middle of the cell. As a matter of
fact, the location of the neck of a Helmholtz resonator has a small influence on
its eigenfrequency. The essential elements are an acoustic mass and an acoustic
compliance, related to the volume. At low frequencies, it is therefore possible
to modify a (symmetrical) T -shaped cell by moving the input and output by
the same length, δ, without changing the transfer matrix (the condition being
δ << λ, if λ is the wavelength). The accuracy of the model based upon this
division is a priori similar to that of the division into symmetrical cells (this
will be discussed more precisely when comparison with measured global cutoff
frequencies will be presented).
The division into asymmetrical cells makes available a supplementary pa-
rameter. It is possible to set a constant value for the eigenfrequency of the
different cells. From the knowledge of the eigenfrequency and the hole mass,
the lengths ℓln + ℓrn of the cells are obtained. The starting point is an initial
value for the length ℓl1 to the left of the first open hole. Therefore, from the
knowledge of ℓln , the length ℓrn on the right of the hole is deduced, then, from
the spacing between two holes, the length ℓln+1 , etc. Depending on the choice

15
 1490

1480

1470
f [Hz]

1460
c

1450

1440
11 equivalent holes
1430
0 0.5 1 1.5 2 2.5 3
lr,1 [mm]

Figure 7: 2nd method: Division of a lattice with 11 equivalent holes into asym-
metrical cells. On the horizontal axis is the semi-cell length to the right ℓr1 .
The region where a solution (i.e. a possible common eigenfrequency) is found is
dashed and the horizontal dotted lines limit the frequency range where a correct
partition exists.

of the initial length ℓl1 , a more or less wide range of possible eigenfrequencies is
found. The result is that a solution exists for fc ∈ [1439, 1478] Hz, as shown in
Fig.7 (there is a continuum of solutions).
Looking at the partition itself, graphed in Fig.8 for fc = 1470 Hz and lr,1 =
1.50mm, it is observed that some cells are very asymmetrical, the borders being
located very close to the middle of a hole (and even within the hole opening).
This seems to be curious and unreal, but formally the transfer matrix of the
whole lattice is identical to this of an acoustically regular one, and we will see
in the final discussion that the results are interesting. The acoustical regularity
can be far from the geometrical one!
The surprise can diminish if we accept that the found lattice is equivalent
to the symmetrized lattice with the same holes and cells, but with holes at the
center of the cells. We checked that the main discrepancies between the asym-
metrical lattice and the symmetrized one occur at frequencies much higher than
cutoff. Nevertheless the relative error is non negligible at very low frequencies,
and this is intuitive: makers know that the shift of the first open tone hole mod-
ifies the first resonance frequencies. Further analytical analysis confirms that
the expected error due to the moving of the holes is larger at low frequencies
than around cutoff. Finally we notice that this symmetrized lattice cannot be
found by the first method, which is not flexible enough.

16
 18
16 14 8 6 5
2120 19 15 11
9 3 1

1 2 3 4 5 6 7 8 9 10 11

200 250 300 350 400 450 500 550

Distance to reed tip [mm]

Figure 8: 2nd method: Illustration of the partition into 11 asymmetrical cells for
fc = 1470 Hz and lr,1 = 1.50mm. The numbers below are the cell numbers from
the input of the instrument; while the numbers on the top are the hole numbers,
as defined in Appendix A. For three cells, two holes have been replaced by an
equivalent hole.

5.2.3 A possible extension of the definition of a local cutoff frequency

A third method of analysis is not based upon any possible division. If two ad-
jacent, symmetrical T -shaped cells have the same eigenfrequency with different
lengths ℓ1 and ℓ2 and different hole masses mh1 and mh2 , Eq. (6) leads to:
ρ
ℓ1 mh1 = ℓ2 mh2 = , (14)
2Skc2
therefore the spacing d = ℓ1 + ℓ2 between the holes satisfies:
 
1 ρ 1 1
d= 2 + (15)
kc 2S mh1 mh2
thus s  
c ρ 1 1
fc = + . (16)
2π 2Sd mh 1 mh2
This quantity can be calculated without a division of the lattice for every pair
of tone holes. It is the eigenfrequency of a resonator of length d, with two
necks corresponding to the holes with a cross section divided by 2. If it is a
constant over the length of a lattice, the lattice is acoustically regular. If it is
not constant, its variation can be regarded as a measure of irregularity. We can
define the frequency given by Eq. (16) as a local cutoff frequency, depending in a
direct way on the dimensions (masses) of two adjacent holes and their distance.
This extend the definition given in the introduction, and the two definitions are
coherent for either geometrically or acoustically regular lattices.
Fig.9 shows the results of the third method. An important feature is the
difference in variation for the acoustic masses and the local cutoff frequencies.
The maximum variation for the square root of the masses is 2.24 while the max-
imum variation for the cutoff frequencies is 1.38. It can be concluded that the
choice of the spacing between the tone holes allows a significant compensation
for the variation of the hole acoustic masses, and a certain acoustical regularity
exists. This is confirmed by a rough inspection of the holes of a real instrument:
the holes appear to be larger as well as their spacing from the input to the bell.

17
 600 1800

Local cutoff frequency [Hz]

Corrected
Acoustic mass [1/m]

500 Not corrected 1700

1600
400
1500
300
1400
200 1300 Corrected
Not corrected
100 1200
1 2 3 4 5 6 7 8 9 10 11 1−2 2−3 3−4 4−5 5−6 6−7 7−8 8−9 10−11
Cell number Cell number

Figure 9: 3rd method. Left figure: Acoustic mass divided by the density mh /ρ,
for 11 equivalent tone holes. Right figure: Local cutoff frequency for a pair of
adjacent holes (Eq. (16)). For both quantities, the uncorrected and corrected
models (see section 5.3) are compared.

Is this effect directly sought by the makers? It is far from evident, and to our
minds this remains an open question.

5.3 Results with corrections

The objective of the present work is to analyze a real lattice in terms of acoustical
regularity, and implies use of a simple model. Actually many details have an
influence on the local cutoff frequencies, but the order of magnitude of this
influence remains small. In order to validate this idea, two types of corrections
have been studied:

• the effect of closed holes. At low frequencies, this effect is due to the
air compressibility, and is proportional to the volume of the cavity of
the closed hole. The volume of the portion of the main tube involved in
Eq. (6) is therefore modified, in accordance with the lumped elements
hypothesis. Fig.9 shows that the correction for the cutoff is small. The
cutoff frequency is lowered by a typical amount of 25 Hz.
• the effect of the cross-section variation of the main tube. We are interested
in the enlargement of the portion of the tube where open holes are present.
This portion is not long. The choice is to describe this enlargement as the
insertion of a change in conicity, just below hole n◦ 3 (see Appendix A),
by using results explained e.g. in Ref. [20]. This change in conicity is
represented by a supplementary shunt mass mcone , which is rather high,
i.e. equivalent to a narrow open hole. It is given by the following formula:
x
mcone = ρ
S
where x is the length of the missing part of the cone, equal to 287mm.
The mass is inserted at 28.4mm from hole n◦ 3. In addition, the masses
of holes n◦ 1 and n◦ 2 need to be multiplied by the ratios S1 /S and S2 /S,
where Si is the cross section of the main tube at the location of hole n◦ i.
Again the correction of the results appears to be very small (see Fig.9).

18
 Concerning the corrections of the results of the 2nd method of analysis (di-
vision into asymmetrical cells), the effect is small as well. The range of possible
common eigenfrequencies becomes slightly narrower and lower.

6 Global and local cutoff frequencies

Using the results obtained in the previous section, it is possible to compare the
theoretical eigenfrequencies and the measured, global cutoff frequencies. It is
reasonable to think that the use of a simplified model does not modify the dis-
cussion written hereafter, because the corrections are very small. Nevertheless
the results take the two kinds of corrections into account.
Fig.10 shows the results of: i) the measurements of the global cutoffs; ii) the
calculation of the possible constant eigenfrequencies using division into asym-
metrical cells (2nd method §5.2.2); iii) the calculation of the local cutoff fre-
quencies (3rd method, §5.2.3). Notice that there are two different types of axes
for the abscissa: for the theoretical results, the numbers correspond to the cell
numbers, while for the experimental ones, the results depend on the fingerings.

Cell number
1−2 2−3 3−4 4−5 5−6 6−7 7−8 8−9 9−10 10−11
Global f , measured
c
Local f by method 3
c
Range local f by method 2
c

1800
Cutoff frequency [Hz]

1700

1600

1500

1400

1300

1200

1100

1000
A4 G#4 G4 F4 E4 D4 C4 A#3 A3 G3 F3
Written note

Figure 10: Comparison between experimentally determined global fc and local

fc obtained using method 2 and 3 for the corrected model. The values of fc for
method 3 are plotted between the tones, because they are based on the acoustic
mass of two subsequent equivalent tone holes. The dashed region shows the
frequency range for the local fc obtained by the 2nd method.

The most evident feature is the satisfactory result of the division into asym-
metrical cells. The constant eigenfrequency obtained by this method coincides
very well with the measured global cutoff for many fingerings. We think that
this is a validation of both the definition of the acoustical regularity and the

19
method of division into asymmetrical cells. Nevertheless we notice that this ex-
cellent agreement is the consequence of a slight overestimation of the two results:
for the theoretical result, as mentioned earlier, the frequency is higher than the
true one, because of the use of approximation (6); for the experimental result, it
has been seen that the practical measurement of the global cutoff overestimates
the true cutoff as well. Otherwise a discrepancy exists for four lower notes. The
explanation has been given earlier, in section 3.2, when studying the effect of
the termination.

Concerning the 3rd method, it gives an order of magnitude of the local cut-
offs, but they are in general 15% higher than the global measured values, at
least at the ends of the considered register. The tendency of the variation looks
rather similar, except for fingerings around the note A3 (we have no interpreta-
tion of this fact). If for a given fingering the global cutoff was determined by the
value of the cell corresponding to the first open hole, the two frequencies should
coincide (notice however that the results of this method concern a pair of tone
holes). It is not the case, but the fact that the tendency is similar (except for
some notes) is in accordance with the hypothesis that the global frequencies are
determined by the local frequencies of the first cells.
We finally remark that the 3rd method seems to be interesting because of its
simplicity, no division being needed. For sake of simplicity, the improvement of
Eq. (16) by taking into account the correction term of Eq. (7) is not discussed
here. The correction depends on the length of the cell, but the irregularity of
the results shown in Fig. 10 is not significantly reduced.

7 Conclusion

A theoretical definition of acoustical regularity is possible, at least at low fre-

quencies, and can be applied to a clarinet. An important degree of acoustical
regularity is found on a clarinet, despite the rather great geometrical irregular-
ity. This explains why the measured cutoff frequencies do not vary very much,
except when the termination effect occurs. Results for another type of clarinet
probably should be rather similar (see especially Ref. [21]).
The acoustical regularity is limited to wavelengths large compared to the
inter-hole spacing. A consequence is that if higher cutoff frequencies exist, e.g.
limiting a new stop band, the previous analysis of regularity cannot be expected
to be relevant for these frequencies. A second stop band seems to exist on both
experimental and numerical results between roughly 2500 Hz and 3000 Hz, the
second cutoff being different from one fingering to another one. This frequency
band was noticed by Wolfe [22]. The understanding of the existence of this band
can be a subject for future investigation: what is sure is that it is far from the
theoretical next cutoff frequency of the periodic lattice studied in section 3.2.
This frequency, called the Bragg frequency in physics, is the half-wavelength
eigenfrequency of a cell without holes, i.e. c/(4ℓ), equal to 5000 Hz, see e.g.
Ref. [8]).
The division of a real lattice as an acoustically regular lattice is not an easy
task. Thanks to an extension of the definition of such a lattice, the 2nd method

20
gives a satisfactory division. Notice that when Benade wrote about the cutoff
frequency of segments (in the sentences cited in the introduction), he did not
explain how this frequency is defined, i.e. how the two first segments are divided.
The present paper does not present a classical comparison between experi-
ment and theory: a precise comparison between theory and experiment could
be sought, especially by taking into account the effect of key pads, the antisym-
metrical effects of tone holes, or the precise geometry of the bell. However the
agreement between measured global cutoff frequencies and the theoretical cutoff
of the acoustically regular lattice built from the real geometry is very good. The
limitation to long wavelength is not problematic: this was not evident, because
though the spacing between holes is small compared to the cutoff wavelength,
the total length of the lattice is not small at all.
The 3rd method is very simple and gives correct orders of magnitude: the
concept is probably rather close to that in the mind of Benade. When qualita-
tively looking at the location and sizes of the tone holes (see Fig.7), the corre-
lation between larger spacings between holes and wider holes roughly appears,
and this is confirmed by the approach of the 3rd method.
It remains to understand the origin of this correlation. Why do the makers
provide an increase of the spacing between holes together with an increase of
their radius? Probably it is related to the search for correct tuning, because
when a hole is moved downstream, it needs to be enlarged for a given tuning.
If this is true, why do the makers enlarge the holes far from the reed? Is it
related to radiation efficiency, or to nonlinear effects, or to the possibilities
of the fingers and the keys? Another topic for future investigation is a more
complete understanding of the effect of the value of the cutoff frequency on the
tone color.
In this paper the question of the bell of the clarinet has not been studied
in a precise way. Its effect is known to ensure a correct tuning of the second
register (see refs [23–25]). An idea could be that for tone-color purposes, the
shape of the bell, which is nearly catenoidal, is sought to be equivalent to a
continuation of the open tone-hole lattice. According to Ref. [23], the cutoff of
the (infinite) catenoidal horn should be f = mc/2π, where m is the expansion
parameter, found to be 1/0.085 m−1 . This would lead to a value of 636 Hz,
much lower than the cutoff of the tone-hole lattice. Therefore this simple idea
is not satisfactory, and anyway it ignores the finite size effect of periodic media.
Finally we remark that an extension of this study to other cylindrical wood-
winds, such as the flute, and even to conical instruments, such as saxophones
or oboes, is possible without great complexity.

APPENDICES

A Geometry of the clarinet studied

The dimensions and locations of the holes of the clarinet Yamaha Y250 are given
in Table 1. Table 2 indicates the first opened holes for the different fingerings.

21
Table 1: Numerical data for the complete set of tone holes. Holes are numbered
for decreasing distance to the tip of the reed. Hole 24 is the register hole.
Holes 22 and 23 are not used for basic fingerings (see Table 2), therefore they
have not been considered in the study. The uncertainties of the measurements
are ±0.02mm for the tone-hole radius, ±0.04mm for the tone-hole height, and
±0.005mm for the tube radius.
no. x [mm] Hole radius b [mm] Bore radius a [mm] Hole height h [mm]
24 156.50 1.50 7.53 12.50
23 167.33 2.46 7.50 6.90
22 193.98 2.88 7.46 6.87
21 203.51 2.70 7.46 6.66
20 213.74 2.47 7.46 6.81
19 231.14 2.47 7.46 6.68
18 239.66 3.58 7.47 10.23
17 241.78 2.48 7.47 6.89
16 252.90 2.69 7.47 8.95
15 271.38 2.36 7.47 6.91
14 285.12 3.44 7.47 9.20
13 287.43 2.75 7.47 6.60
12 289.95 2.83 7.47 6.67
11 308.62 3.94 7.47 7.19
10 318.88 2.61 7.47 6.66
9 349.86 3.65 7.49 6.08
8 365.19 4.13 7.51 8.72
7 370.36 3.80 7.52 6.99
6 391.39 4.02 7.54 8.66
5 414.34 4.91 7.55 8.66
4 446.63 5.25 7.53 5.13
3 473.60 6.22 7.52 5.27
2 505.78 5.70 7.67 4.63
1 544.10 5.71 8.40 4.46

22
Table 2: Most common notes of a B♭ clarinet (written pitches). The third
column gives the target frequency of a well tuned note when it is normally
tempered. The fourth column gives the hole number of the first open hole(s) in
the lattice. When no number is listed, this note is not considered in this study
and does not belong to the “normal” set of fingerings. Excluding a note from
this set means that for its corresponding fingering, it is necessary to open a tone
hole that keeps closed for the other fingerings.
Note fp [Hz] First opened tone-hole number
Chalumeau E3 147 all closed
F3 156 1
F♯3 165 -
G3 175 3
G♯3 185 -
A3 196 5
A♯3 208 6
B3 220 -
C4 233 8+9
C♯4 247 -
D4 262 11
D♯4 277 -
E4 294 14+15
F4 311 16
F♯4 330 -
Throat G4 349 18+19
G♯4 370 20
A4 392 21
A♯4 415 -
Clarinet B4 440 all closed
C5 466 1
C♯5 494 -
D5 523 3
D♯5 554 -
E5 587 5
F5 622 6
F♯5 659 -
G5 698 8+9
G♯5 740 -
A5 784 11
A♯5 831 -
B5 880 14+15
C6 932 16

23
B Use of a more complete model
We have ignored the negative acoustic masses corresponding to the antisymmet-
rical field in the holes. If the series impedance Za = jωma is taken into account
together with the shunt admittance Y = (jωmh )−1 , the transfer matrix of a
hole is written as follows (see e.g Fig.3 of Ref. [7]):
 
1 1 + Za Y /4 Za
.
1 − Za Y /4 Y 1 + Za Y /4

Multiplying this matrix on both sides by the transfer matrix of a segment of

cylindrical tube of length ℓ leads to the coefficients of the matrix of the T-shaped
cell:
 
Y
B = cos kℓ + j zc sin kℓ (Za cos kℓ + 2jzc sin kℓ) (1 − Za Y /4)−1
2
 

Za −1 −1
C = Y cos kℓ + 2jzc sin kℓ cos kℓ + j zc sin kℓ (1 − Za Y /4) .
−1
2

This result exhibits the four types of cutoff frequencies, the lowest one corre-
sponding to the vanishing of the first factor of the coefficient C. As expected,
all of them depend on either the series impedance or the shunt admittance, for
reasons of symmetry. Therefore the exact value of the cutoffs are simpler than
those obtained after approximations, as has been done in Ref. [9].
In order to study the acoustical regularity, Eq. (10) is transformed into:

B 1 + kkc tan kc ℓ tan kℓ 1 − j Z2a cot kℓ

Zc = = zc2 .
C 1 − kkc tan kc ℓ cot kℓ 1 + j Z2a tan kℓ

At low frequencies, the characteristic impedance given by Eq. (10) is multi-

plied by the factor (1 + ma /2m)1/2 . As a consequence, it is possible to ensure
an improved acoustical regularity, as follows: in order to obtain a constant
characteristic impedance at every frequency, both the cutoff fc and the ratio
(1 + ma /2m)/S 2 can be chosen to be equal. Therefore, for a given length ℓ,
there are two equations for the two parameters of the holes (height h and radius
b).
However the ratio ma /2m ≃ −0.18b3/(a2 ℓ) is in general close to 0.01 or
0.02, thus it is not important to take this term into account in the present
study, because a high precision is not needed.

References
[1] A.H. Benade, On the mathematical theory of woodwind finger holes, J.
Acoust. Soc. Am. 32(1960), 1591-1608.
[2] A. H. Benade: Fundamentals of Musical Acoustics. Oxford University
Press, London, 1976.
[3] A.H. Benade and D.H. Keefe. The physics of a new clarinet design. The
Galpin Society Journal XLIX (1996), 113-142.

24
 [4] J. Wolfe and J. Smith : Cutoff frequencies and cross fingerings in baroque,
classical, and modern flutes. J. Acoust. Soc. Am. 114 (2003), 2263-2272.
[5] A. Chaigne, J. Kergomard: Acoustique des instruments de musique. Belin,
Paris, 2008.
[6] D. H. Keefe : Theory on the single woodwind tone hole. J. Acoust. Soc.
Am., 72(1982), 676-687.
[7] V. Dubos, J. Kergomard, A. Khettabi, J. P. Dalmont, D. H. Keefe et C.J
Nederveen: Theory of sound propagation in a duct with a branched tube
using modal decomposition. Acustica Acta Acustica, 85 (1999), 153-169.
[8] J. Kergomard: Champ interne et champ externe des instruments à vent.
Thèse de Doctorat d’Etat, Université Pierre et Marie Curie, 1981.
[9] D. H. Keefe: Woodwind air column models. J. Acoust. Soc. Am., 88 (1990),
35-51.
[10] J. Kergomard: Tone hole external interactions in woodwinds musical in-
struments, 13th International Congress on Acoustics, Yugoslavia, 53-56,
1989.
[11] A. H. Benade, S. N. Kouzoupis: The clarinet spectrum : theory and exper-
iment. J. Acoust. Soc. Am., 83(1988), 292-304.
[12] A.H. Benade. Wind instruments and music acoustics. In “Sound Generation
in Winds, Strings, Computers”, No. 29, pages 14-99 (1980),Royal Swedish
Academy of Music.
[13] J. Kergomard, E. Moers, J.-D. Dalmont, Ph. Guillemain: Cutoff frequen-
cies of a clarinet. What is acoustical regularity? Forum Acusticum 2011,
533-538, Aalborg, Denmark.
[14] C.A. Macaluso, J.-P. Dalmont, Trumpet with near-perfect harmonicity:
Design and acoustic results. J. Acoust. Soc. Am. Volume 129 (2011), 404-
414.
[15] P.W. Anderson:Absence of diffusion in certain random lattices. Phys. Rev.
109 (1958), 1492-1505.
[16] H. Fürstenberg: Noncommuting random matrix products, Trans. Am.
Math. Soc. 108 (1963), 377-428.
[17] C. Depollier, J. Kergomard, F. Laloë, Localisation d’Anderson des on-
des dans les réseaux acoustiques unidimensionnels aléatoires, Annales de
physique 11 (1986), 457-492.
[18] A. H. Benade: Physics of Brasses, Scientific American, July 1973, 24-35.
[19] E.M.T. Moers: Acoustical analysis of the open tone hole lattice of a clarinet.
Master Report, Technische Universiteit Eindhoven, Department of Applied
Physics, June 2010.
[20] A. H. Benade: Equivalent circuits for conical waveguides. J. Acoust. Soc.
Am., 88 (1988):1764-1769.

25
[21] University of New South Wales, http://www.phys.unsw.edu.au/music/clarinet
[22] University of New South Wales, http://www.phys.unsw.edu.au/jw/cutoff.html
[23] C.J. Nederveen: Acoustical aspects of woodwind instruments. Northern
Illinois University press, 1969.
[24] C.J. Nederveen, J.P. Dalmont: Corrections to the plane-wave approxima-
tion in rapidly flaring horns, Acta Acustica united with Acustica, 94 (2008),
461-173.
[25] V. Debut, J. Kergomard, F. Laloë, Analysis and optimisation of the tuning
of the twelfths for a clarinet resonator, Applied Acoustics 66 (2005), 365-
409.

Acknowledgments
We thank Didier Ferrand and Alain Busso for their help for the experiments,
and René Caussé, Jean-Pierre Dalmont, Douglas Keefe, Philippe Guillemain,
Avraham Hirschberg, Franck Laloë, Kees Nederveen, P.-A. Taillard and Joe
Wolfe for useful discussions. We thank also the referees and Murray Campbell
for their help in improving the manuscript.

26